Brave takes on ChatGPT with private AI assistant Leo

Every major browser wants to be your AI copilot right now, but almost all of them demand a login and quietly stash your queries for future model training.

Privacy-focused browser Brave is taking a very different route. Following months of testing, the company is rolling out Leo, a built-in AI assistant for all desktop users running version 1.60. The rollout is happening in phases over the next few days, with iOS and Android apps scheduled to get the feature in the coming months.

On the surface, Leo handles the standard assistant playbook: translating text, answering random questions, summarizing long web pages, and drafting content. But the real difference is how it handles your data. You do not need an account to use it, your chats are never recorded, and nothing you type is used to train AI models.

Under the hood, the free tier runs Meta's open-source Llama 2 model. If you want more horsepower, Brave is offering Leo Premium for $15 a month, which unlocks Anthropic's faster Claude Instant model, higher rate limits, priority access during peak hours, and early access to new features.

Brave CTO Brian Bondy noted that Leo's architecture is designed to be model-agnostic, meaning the browser plans to plug in more models down the road so users can choose what works best for them.

Why it matters

We think Brave found the exact weak spot in Big Tech's AI strategy: trust.

For everyday users and privacy-conscious pros, Leo is an immediately appealing utility. If you are reviewing confidential internal memos, legal briefs, or sensitive medical pages, pasting that text into Bing Chat or ChatGPT feels risky. Leo offers a clean, anonymous sandbox right inside the browser where you can summarize confidential tabs without wondering where your data ends up.

The friction, however, is the price tag. At $15 a month, Leo Premium asks a lot for Claude Instant, especially when $20 gets you frontier models like GPT-4 elsewhere. The free Llama 2 tier is great for basic page summaries, but power users will likely hit its reasoning ceiling pretty quickly.

Still, Brave's approach proves that browser AI doesn't have to be an aggressive data-harvesting machine. By treating AI models as swappable backends while guaranteeing absolute privacy, Brave is giving users a genuine alternative to the surveillance-heavy giants.